lustre/llite/vvp_dev.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * cl_device and cl_device_type implementation for VVP layer.
  37  *
  38  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  39  *   Author: Jinshan Xiong <jinshan.xiong@intel.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_LLITE
  43
  44 #ifndef __KERNEL__
  45 # error This file is kernel only.
  46 #endif
  47
  48 #include <obd.h>
  49 #include <lustre_lite.h>
  50
  51 #include "vvp_internal.h"
  52
  53 /*****************************************************************************
  54  *
  55  * Vvp device and device type functions.
  56  *
  57  */
  58
  59 /*
  60  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  61  * "llite_" (var. "ll_") prefix.
  62  */
  63
  64 struct kmem_cache *vvp_thread_kmem;
  65 static struct kmem_cache *vvp_session_kmem;
  66 static struct lu_kmem_descr vvp_caches[] = {
  67         {
  68                 .ckd_cache = &vvp_thread_kmem,
  69                 .ckd_name  = "vvp_thread_kmem",
  70                 .ckd_size  = sizeof (struct vvp_thread_info),
  71         },
  72         {
  73                 .ckd_cache = &vvp_session_kmem,
  74                 .ckd_name  = "vvp_session_kmem",
  75                 .ckd_size  = sizeof (struct vvp_session)
  76         },
  77         {
  78                 .ckd_cache = NULL
  79         }
  80 };
  81
  82 static void *vvp_key_init(const struct lu_context *ctx,
  83                           struct lu_context_key *key)
  84 {
  85         struct vvp_thread_info *info;
  86
  87         OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, __GFP_IO);
  88         if (info == NULL)
  89                 info = ERR_PTR(-ENOMEM);
  90         return info;
  91 }
  92
  93 static void vvp_key_fini(const struct lu_context *ctx,
  94                          struct lu_context_key *key, void *data)
  95 {
  96         struct vvp_thread_info *info = data;
  97         OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
  98 }
  99
 100 static void *vvp_session_key_init(const struct lu_context *ctx,
 101                                   struct lu_context_key *key)
 102 {
 103         struct vvp_session *session;
 104
 105         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, __GFP_IO);
 106         if (session == NULL)
 107                 session = ERR_PTR(-ENOMEM);
 108         return session;
 109 }
 110
 111 static void vvp_session_key_fini(const struct lu_context *ctx,
 112                                  struct lu_context_key *key, void *data)
 113 {
 114         struct vvp_session *session = data;
 115         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 116 }
 117
 118
 119 struct lu_context_key vvp_key = {
 120         .lct_tags = LCT_CL_THREAD,
 121         .lct_init = vvp_key_init,
 122         .lct_fini = vvp_key_fini
 123 };
 124
 125 struct lu_context_key vvp_session_key = {
 126         .lct_tags = LCT_SESSION,
 127         .lct_init = vvp_session_key_init,
 128         .lct_fini = vvp_session_key_fini
 129 };
 130
 131 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 132 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
 133
 134 static const struct lu_device_operations vvp_lu_ops = {
 135         .ldo_object_alloc      = vvp_object_alloc
 136 };
 137
 138 static const struct cl_device_operations vvp_cl_ops = {
 139         .cdo_req_init = ccc_req_init
 140 };
 141
 142 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 143                                           struct lu_device_type *t,
 144                                           struct lustre_cfg *cfg)
 145 {
 146         return ccc_device_alloc(env, t, cfg, &vvp_lu_ops, &vvp_cl_ops);
 147 }
 148
 149 static const struct lu_device_type_operations vvp_device_type_ops = {
 150         .ldto_init = vvp_type_init,
 151         .ldto_fini = vvp_type_fini,
 152
 153         .ldto_start = vvp_type_start,
 154         .ldto_stop  = vvp_type_stop,
 155
 156         .ldto_device_alloc = vvp_device_alloc,
 157         .ldto_device_free  = ccc_device_free,
 158         .ldto_device_init  = ccc_device_init,
 159         .ldto_device_fini  = ccc_device_fini
 160 };
 161
 162 struct lu_device_type vvp_device_type = {
 163         .ldt_tags     = LU_DEVICE_CL,
 164         .ldt_name     = LUSTRE_VVP_NAME,
 165         .ldt_ops      = &vvp_device_type_ops,
 166         .ldt_ctx_tags = LCT_CL_THREAD
 167 };
 168
 169 /**
 170  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 171  * pressure, when environments cannot be allocated.
 172  */
 173 int vvp_global_init(void)
 174 {
 175         int result;
 176
 177         result = lu_kmem_init(vvp_caches);
 178         if (result == 0) {
 179                 result = ccc_global_init(&vvp_device_type);
 180                 if (result != 0)
 181                         lu_kmem_fini(vvp_caches);
 182         }
 183         return result;
 184 }
 185
 186 void vvp_global_fini(void)
 187 {
 188         ccc_global_fini(&vvp_device_type);
 189         lu_kmem_fini(vvp_caches);
 190 }
 191
 192
 193 /*****************************************************************************
 194  *
 195  * mirror obd-devices into cl devices.
 196  *
 197  */
 198
 199 int cl_sb_init(struct super_block *sb)
 200 {
 201         struct ll_sb_info *sbi;
 202         struct cl_device  *cl;
 203         struct lu_env     *env;
 204         int rc = 0;
 205         int refcheck;
 206
 207         sbi  = ll_s2sbi(sb);
 208         env = cl_env_get(&refcheck);
 209         if (!IS_ERR(env)) {
 210                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 211                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 212                 if (!IS_ERR(cl)) {
 213                         cl2ccc_dev(cl)->cdv_sb = sb;
 214                         sbi->ll_cl = cl;
 215                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 216                 }
 217                 cl_env_put(env, &refcheck);
 218         } else
 219                 rc = PTR_ERR(env);
 220         RETURN(rc);
 221 }
 222
 223 int cl_sb_fini(struct super_block *sb)
 224 {
 225         struct ll_sb_info *sbi;
 226         struct lu_env     *env;
 227         struct cl_device  *cld;
 228         int                refcheck;
 229         int                result;
 230
 231         ENTRY;
 232         sbi = ll_s2sbi(sb);
 233         env = cl_env_get(&refcheck);
 234         if (!IS_ERR(env)) {
 235                 cld = sbi->ll_cl;
 236
 237                 if (cld != NULL) {
 238                         cl_stack_fini(env, cld);
 239                         sbi->ll_cl = NULL;
 240                         sbi->ll_site = NULL;
 241                 }
 242                 cl_env_put(env, &refcheck);
 243                 result = 0;
 244         } else {
 245                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 246                 result = PTR_ERR(env);
 247         }
 248         /*
 249          * If mount failed (sbi->ll_cl == NULL), and this there are no other
 250          * mounts, stop device types manually (this usually happens
 251          * automatically when last device is destroyed).
 252          */
 253         lu_types_stop();
 254         RETURN(result);
 255 }
 256
 257 /****************************************************************************
 258  *
 259  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 260  *
 261  ****************************************************************************/
 262
 263 /*
 264  * To represent contents of a page cache as a byte stream, following
 265  * information if encoded in 64bit offset:
 266  *
 267  *       - file hash bucket in lu_site::ls_hash[]       28bits
 268  *
 269  *       - how far file is from bucket head              4bits
 270  *
 271  *       - page index                                   32bits
 272  *
 273  * First two data identify a file in the cache uniquely.
 274  */
 275
 276 #define PGC_OBJ_SHIFT (32 + 4)
 277 #define PGC_DEPTH_SHIFT (32)
 278
 279 struct vvp_pgcache_id {
 280         unsigned                 vpi_bucket;
 281         unsigned                 vpi_depth;
 282         uint32_t                 vpi_index;
 283
 284         unsigned                 vpi_curdep;
 285         struct lu_object_header *vpi_obj;
 286 };
 287
 288 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 289 {
 290         CLASSERT(sizeof(pos) == sizeof(__u64));
 291
 292         id->vpi_index  = pos & 0xffffffff;
 293         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 294         id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
 295 }
 296
 297 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 298 {
 299         return
 300                 ((__u64)id->vpi_index) |
 301                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 302                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 303 }
 304
 305 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 306                                cfs_hlist_node_t *hnode, void *data)
 307 {
 308         struct vvp_pgcache_id   *id  = data;
 309         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 310
 311         if (id->vpi_curdep-- > 0)
 312                 return 0; /* continue */
 313
 314         if (lu_object_is_dying(hdr))
 315                 return 1;
 316
 317         cfs_hash_get(hs, hnode);
 318         id->vpi_obj = hdr;
 319         return 1;
 320 }
 321
 322 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 323                                          struct lu_device *dev,
 324                                          struct vvp_pgcache_id *id)
 325 {
 326         LASSERT(lu_device_is_cl(dev));
 327
 328         id->vpi_depth &= 0xf;
 329         id->vpi_obj    = NULL;
 330         id->vpi_curdep = id->vpi_depth;
 331
 332         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 333                                 vvp_pgcache_obj_get, id);
 334         if (id->vpi_obj != NULL) {
 335                 struct lu_object *lu_obj;
 336
 337                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 338                 if (lu_obj != NULL) {
 339                         lu_object_ref_add(lu_obj, "dump", current);
 340                         return lu2cl(lu_obj);
 341                 }
 342                 lu_object_put(env, lu_object_top(id->vpi_obj));
 343
 344         } else if (id->vpi_curdep > 0) {
 345                 id->vpi_depth = 0xf;
 346         }
 347         return NULL;
 348 }
 349
 350 static loff_t vvp_pgcache_find(const struct lu_env *env,
 351                                struct lu_device *dev, loff_t pos)
 352 {
 353         struct cl_object     *clob;
 354         struct lu_site       *site;
 355         struct vvp_pgcache_id id;
 356
 357         site = dev->ld_site;
 358         vvp_pgcache_id_unpack(pos, &id);
 359
 360         while (1) {
 361                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 362                         return ~0ULL;
 363                 clob = vvp_pgcache_obj(env, dev, &id);
 364                 if (clob != NULL) {
 365                         struct inode *inode = ccc_object_inode(clob);
 366                         struct page *vmpage;
 367                         int nr;
 368
 369                         nr = find_get_pages_contig(inode->i_mapping,
 370                                                    id.vpi_index, 1, &vmpage);
 371                         if (nr > 0) {
 372                                 id.vpi_index = vmpage->index;
 373                                 /* Cant support over 16T file */
 374                                 nr = !(vmpage->index > 0xffffffff);
 375                                 page_cache_release(vmpage);
 376                         }
 377
 378                         lu_object_ref_del(&clob->co_lu, "dump", current);
 379                         cl_object_put(env, clob);
 380                         if (nr > 0)
 381                                 return vvp_pgcache_id_pack(&id);
 382                 }
 383                 /* to the next object. */
 384                 ++id.vpi_depth;
 385                 id.vpi_depth &= 0xf;
 386                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 387                         return ~0ULL;
 388                 id.vpi_index = 0;
 389         }
 390 }
 391
 392 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 393         if (test_bit(PG_##flag, &(page)->flags)) {                  \
 394                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 395                 has_flags = 1;                                          \
 396         }                                                               \
 397 } while(0)
 398
 399 static void vvp_pgcache_page_show(const struct lu_env *env,
 400                                   struct seq_file *seq, struct cl_page *page)
 401 {
 402         struct ccc_page *cpg;
 403         struct page      *vmpage;
 404         int              has_flags;
 405
 406         cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
 407         vmpage = cpg->cpg_page;
 408         seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
 409                    0 /* gen */,
 410                    cpg, page,
 411                    "none",
 412                    cpg->cpg_write_queued ? "wq" : "- ",
 413                    cpg->cpg_defer_uptodate ? "du" : "- ",
 414                    PageWriteback(vmpage) ? "wb" : "-",
 415                    vmpage,
 416                    PFID(ll_inode2fid(vmpage->mapping->host)),
 417                    vmpage->mapping->host, vmpage->index,
 418                    page_count(vmpage));
 419         has_flags = 0;
 420         seq_page_flag(seq, vmpage, locked, has_flags);
 421         seq_page_flag(seq, vmpage, error, has_flags);
 422         seq_page_flag(seq, vmpage, referenced, has_flags);
 423         seq_page_flag(seq, vmpage, uptodate, has_flags);
 424         seq_page_flag(seq, vmpage, dirty, has_flags);
 425         seq_page_flag(seq, vmpage, writeback, has_flags);
 426         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 427 }
 428
 429 static int vvp_pgcache_show(struct seq_file *f, void *v)
 430 {
 431         loff_t                   pos;
 432         struct ll_sb_info       *sbi;
 433         struct cl_object        *clob;
 434         struct lu_env           *env;
 435         struct vvp_pgcache_id    id;
 436         int                      refcheck;
 437         int                      result;
 438
 439         env = cl_env_get(&refcheck);
 440         if (!IS_ERR(env)) {
 441                 pos = *(loff_t *) v;
 442                 vvp_pgcache_id_unpack(pos, &id);
 443                 sbi = f->private;
 444                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 445                 if (clob != NULL) {
 446                         struct inode *inode = ccc_object_inode(clob);
 447                         struct cl_page *page = NULL;
 448                         struct page *vmpage;
 449
 450                         result = find_get_pages_contig(inode->i_mapping,
 451                                                       id.vpi_index, 1, &vmpage);
 452                         if (result > 0) {
 453                                 lock_page(vmpage);
 454                                 page = cl_vmpage_page(vmpage, clob);
 455                                 unlock_page(vmpage);
 456
 457                                 page_cache_release(vmpage);
 458                         }
 459
 460                         seq_printf(f, "%8x@"DFID": ", id.vpi_index,
 461                                    PFID(lu_object_fid(&clob->co_lu)));
 462                         if (page != NULL) {
 463                                 vvp_pgcache_page_show(env, f, page);
 464                                 cl_page_put(env, page);
 465                         } else
 466                                 seq_puts(f, "missing\n");
 467                         lu_object_ref_del(&clob->co_lu, "dump", current);
 468                         cl_object_put(env, clob);
 469                 } else
 470                         seq_printf(f, "%llx missing\n", pos);
 471                 cl_env_put(env, &refcheck);
 472                 result = 0;
 473         } else
 474                 result = PTR_ERR(env);
 475         return result;
 476 }
 477
 478 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 479 {
 480         struct ll_sb_info *sbi;
 481         struct lu_env     *env;
 482         int                refcheck;
 483
 484         sbi = f->private;
 485
 486         env = cl_env_get(&refcheck);
 487         if (!IS_ERR(env)) {
 488                 sbi = f->private;
 489                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 490                         pos = ERR_PTR(-EFBIG);
 491                 else {
 492                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 493                                                 *pos);
 494                         if (*pos == ~0ULL)
 495                                 pos = NULL;
 496                 }
 497                 cl_env_put(env, &refcheck);
 498         }
 499         return pos;
 500 }
 501
 502 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 503 {
 504         struct ll_sb_info *sbi;
 505         struct lu_env     *env;
 506         int                refcheck;
 507
 508         env = cl_env_get(&refcheck);
 509         if (!IS_ERR(env)) {
 510                 sbi = f->private;
 511                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 512                 if (*pos == ~0ULL)
 513                         pos = NULL;
 514                 cl_env_put(env, &refcheck);
 515         }
 516         return pos;
 517 }
 518
 519 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 520 {
 521         /* Nothing to do */
 522 }
 523
 524 static struct seq_operations vvp_pgcache_ops = {
 525         .start = vvp_pgcache_start,
 526         .next  = vvp_pgcache_next,
 527         .stop  = vvp_pgcache_stop,
 528         .show  = vvp_pgcache_show
 529 };
 530
 531 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 532 {
 533         struct ll_sb_info       *sbi = PDE_DATA(inode);
 534         struct seq_file         *seq;
 535         int                     result;
 536
 537         result = seq_open(filp, &vvp_pgcache_ops);
 538         if (result == 0) {
 539                 seq = filp->private_data;
 540                 seq->private = sbi;
 541         }
 542         return result;
 543 }
 544
 545 struct file_operations vvp_dump_pgcache_file_ops = {
 546         .owner   = THIS_MODULE,
 547         .open    = vvp_dump_pgcache_seq_open,
 548         .read    = seq_read,
 549         .llseek  = seq_lseek,
 550         .release = seq_release,
 551 };